Cloud platforms
Lane A uses models you already buy. Lane B runs in your VPC.
Lane A: models you already buy
Models you already buy would connect with your own credentials, the same way as any other provider. Replay and switching would use them only while they stay enabled. That connection is not running. The same open-weight model can have a different price on each host. Parezen is designed to keep those prices versioned and to prefer the cheaper healthy host when your policy allows it. That host switching is not built. Marketplace listings are not a current build.
Lane B: gateway in your VPC
You would run the gateway next to the workloads, on the platform you already use. Likely targets include ECS, Cloud Run, EKS and GKE. Those names are examples of where teams ask to run it, not a promise that each one has a one-click install.
Your app still uses https://api.parezen.com/v1 only if you choose the hosted control plane. A private deploy uses your own base URL. Tagging headers are the same.